Re: [Apbs-users] nonuniform mesh?

[Gernot K. <ge...@ch...>]

Hi!

On Wed, 2010-08-25 at 11:57 -0500, Daniel Ensign wrote:
> Easily improved would be like heavenly manna. :) I have attached an
> input file for one of the tests I'm trying, along with a relevant pqr. I
> do not know if I need to be so conservative as to do three
> calculations, or whether I need to do several more layers. Briefly,
> here's what I'm doing, using mg-manual:

Attached find a modified simple-example.in including some comments.
Total memory usage in my case was 552.724 MB high water instead of 2.5
GB as with the old script.

There were two issues:

1) Your second grid did not contain all atoms of the molecule, which
might have caused the "unsually large potential values" warnings.

In a three-step focusing procedure the large grid can be rather large
and of low resolution as the solution is only used for the boundary of
the next grid. This way one can also use the faster "bcfl sdh" boundary
condition. 

The second grid should then contain.

2) 0.05 Angstroms is quite a high resolution. I would first try whether
you could live with 0.1 Angstroms. This modification saved 2 GB of
memory on my system!

3) In the last focusing step you used "chgm spl0" (trilinear
interpolation) instead of "chgm spl4" (spline interpolation). I think
that's bad because you charge distribution is inconsistent between the runs.

However, I still got warnings like this 

"% C_VEC: trapped exp overflows:           14"

and

"Vpmg_qmEnergy:  Chopped EXP 158 times".

I also get

"Total electrostatic energy = -1.543597540833E+32 kJ/mol"

I'm not sure why. But the potential itself looks reasonable.

Regards,
Gernot